Three-dimensional Tensor Controlled-Source Audio-Frequency Magnetotelluric Inversion Using LBFGS

The controlled-source audio-frequency magnetotelluric (CSAMT) method has become an important method in geophysical electromagnetic exploration. However, traditional CSAMT only gathers a single set of orthogonal electric and magnetic data, which cannot describe the whole subsurface geological structure. Due to increasingly complex geological targets, the drawbacks of traditional CSAMT have gradually become more significant, promoting the need for tensor CSAMT. Tensor CSAMT can gather richer information, but the 3D forward and inversion models of this method have developed slowly since it was first proposed. The common method for inverting the data of the tensor CSAMT is still magnetotelluric (MT). This paper adopts a staggered-grid finite difference method to realise the 3D forward modelling of the tensor CSAMT. On this basis, we adopt a limited-memory Broyden-Fletcher-Goldfarb-Shanno (LBFGS) method to implement a 3D inversion with full impedance data. Through inverting synthetic and real data, we prove that: (1) directly using an MT method to invert the data of the tensor CSAMT will obtain an incorrect result, (2) the inversion result of tensor CSAMT is more reliable than that of the traditional CSAMT, and (3) LBFGS is more efficient than the nonlinear conjugate gradient (NLCG) for tensor CSAMT. Our research shows that 3D tensor CSAMT inversion with LBFGS is very useful and practical for electromagnetic exploration.